In my much younger days I was much enamored by simple neural networks
and once built a Perceptron (Rosenblatt, 1958) in software. Simple
neural nets were on the out amongst aritificial intelligence researchers
in the early 70's when I built mine but I was still fascinated by what
they could do vs. what they couldn't. Mine was really simple and was
only used to do character recognition of block letters. One of the
things that inspired me was work done at Stanford on a very similar
machine called an Adaline. The Adaline was used to predict rainfall
(yes/no) for the San Francisco area based *only* on barometric pressure
as reported from many local area stations for the day preceding the
prediction. It was actually slightly more accurate (87 vs 85% if I
recall correctly) than the official weather bureau forecast.
It makes me wonder what could be achieved by a really large array of
barometric reporting stations and a single modern PC. The software for
something like a Perceptron or Adaline is pretty simple.
Chuck Norcutt
Sandy Harris wrote:
> Frank van Lindert <Frank.van@xxxxxxxxxxx> wrote:
>
>> Davis is considered a very good brand of weather station for amateur
>> to private semi-professional use, e.g. in agriculture. ...
>
>> But the quality applies only for measuring, not for predicting the
>> weather! No single weather station can do that appropriately.
>
> Would sheer numbers help much? There's a lot of connectivity
> options these days. If someone came up with a really cheap
> basic instrument pack and deployed thousands of them (maybe
> one on every farm?) with communications to some central
> computer, would prediction get any better?
>
> You don't need much bandwidth. Temperature over a 200
> degree range with 1/10 degree accuracy takes 11 bits
> A 128-bit packet would likely do everything you could
> possibly want; an ordinary phone line or cell phone
> connection can handle 512 of those a second.
>
> Designing the cheap weather station might be hard.
> Getting the receiving computer to make sense of all
> that data almost certainly would be.
>
> But if someone did all that, would there be much
> payoff? Or is the problem in forecasting something
> else, so more data would make little difference.
>
==============================================
List usage info: http://www.zuikoholic.com
List nannies: olympusadmin@xxxxxxxxxx
==============================================
|